Archives

The reading map for a build-first AI-native engineering series: nine focused posts, one capstone, and a rigorous learner-critic release bar.

Turn a backlog into a safe, observable conveyor belt where agents make incremental progress.

Split independent work across isolated agents while controlling cost, conflict, and retries.

Reduce blast radius with boundaries, least privilege, approvals, and observable failure paths.

Review behavior, tests, traces, and risk—not just a plausible-looking diff.

Give a fresh agent the map, invariants, commands, and handoffs it needs to work safely.

Govern discovery, authorization, routing, and audit for Model Context Protocol tools.

Curate the smallest high-signal context and preserve state across long agent runs.

Turn team know-how into discoverable, versioned, testable Agent Skills.

Decide when a deterministic function, workflow, or bounded agent is the right tool.

A practical Go backend blueprint for AI workflows and real-time products with clear service boundaries, queues, idempotency, observability, and Kubernetes delivery.

How to design a useful LLM evaluation set for production workflows using real task cases, failure modes, regression checks, and human review.

A practical checklist for turning an AI feature from a happy-path demo into a system people can trust.

Why this site has a blog, and what I intend to put here.